1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $Id$ 35 */ 36 37 #ifndef TUBA_INCLUDE 38 #include <sys/param.h> 39 #include <sys/queue.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/sysctl.h> 43 #include <sys/malloc.h> 44 #include <sys/mbuf.h> 45 #include <sys/protosw.h> 46 #include <sys/socket.h> 47 #include <sys/socketvar.h> 48 #include <sys/errno.h> 49 #include <sys/syslog.h> 50 51 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 52 53 #include <net/if.h> 54 #include <net/route.h> 55 56 #include <netinet/in.h> 57 #include <netinet/in_systm.h> 58 #include <netinet/ip.h> 59 #include <netinet/in_pcb.h> 60 #include <netinet/ip_var.h> 61 #include <netinet/tcp.h> 62 #include <netinet/tcp_fsm.h> 63 #include <netinet/tcp_seq.h> 64 #include <netinet/tcp_timer.h> 65 #include <netinet/tcp_var.h> 66 #include <netinet/tcpip.h> 67 #ifdef TCPDEBUG 68 #include <netinet/tcp_debug.h> 69 static struct tcpiphdr tcp_saveti; 70 #endif 71 72 static int tcprexmtthresh = 3; 73 tcp_seq tcp_iss; 74 tcp_cc tcp_ccgen; 75 76 struct tcpstat tcpstat; 77 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 78 CTLFLAG_RD, &tcpstat , tcpstat, ""); 79 80 static int log_in_vain = 0; 81 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 82 &log_in_vain, 0, ""); 83 84 u_long tcp_now; 85 struct inpcbhead tcb; 86 struct inpcbinfo tcbinfo; 87 88 static void tcp_dooptions __P((struct tcpcb *, 89 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 90 static void tcp_pulloutofband __P((struct socket *, 91 struct tcpiphdr *, struct mbuf *)); 92 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 93 static void tcp_xmit_timer __P((struct tcpcb *, int)); 94 95 #endif /* TUBA_INCLUDE */ 96 97 /* 98 * Insert segment ti into reassembly queue of tcp with 99 * control block tp. Return TH_FIN if reassembly now includes 100 * a segment with FIN. The macro form does the common case inline 101 * (segment is the next to be received on an established connection, 102 * and the queue is empty), avoiding linkage into and removal 103 * from the queue and repetition of various conversions. 104 * Set DELACK for segments received in order, but ack immediately 105 * when segments are out of order (so fast retransmit can work). 106 */ 107 #ifdef TCP_ACK_HACK 108 #define TCP_REASS(tp, ti, m, so, flags) { \ 109 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 110 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 111 (tp)->t_state == TCPS_ESTABLISHED) { \ 112 if (ti->ti_flags & TH_PUSH) \ 113 tp->t_flags |= TF_ACKNOW; \ 114 else \ 115 tp->t_flags |= TF_DELACK; \ 116 (tp)->rcv_nxt += (ti)->ti_len; \ 117 flags = (ti)->ti_flags & TH_FIN; \ 118 tcpstat.tcps_rcvpack++;\ 119 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 120 sbappend(&(so)->so_rcv, (m)); \ 121 sorwakeup(so); \ 122 } else { \ 123 (flags) = tcp_reass((tp), (ti), (m)); \ 124 tp->t_flags |= TF_ACKNOW; \ 125 } \ 126 } 127 #else 128 #define TCP_REASS(tp, ti, m, so, flags) { \ 129 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 130 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 131 (tp)->t_state == TCPS_ESTABLISHED) { \ 132 tp->t_flags |= TF_DELACK; \ 133 (tp)->rcv_nxt += (ti)->ti_len; \ 134 flags = (ti)->ti_flags & TH_FIN; \ 135 tcpstat.tcps_rcvpack++;\ 136 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 137 sbappend(&(so)->so_rcv, (m)); \ 138 sorwakeup(so); \ 139 } else { \ 140 (flags) = tcp_reass((tp), (ti), (m)); \ 141 tp->t_flags |= TF_ACKNOW; \ 142 } \ 143 } 144 #endif 145 #ifndef TUBA_INCLUDE 146 147 static int 148 tcp_reass(tp, ti, m) 149 register struct tcpcb *tp; 150 register struct tcpiphdr *ti; 151 struct mbuf *m; 152 { 153 register struct tcpiphdr *q; 154 struct socket *so = tp->t_inpcb->inp_socket; 155 int flags; 156 157 /* 158 * Call with ti==0 after become established to 159 * force pre-ESTABLISHED data up to user socket. 160 */ 161 if (ti == 0) 162 goto present; 163 164 /* 165 * Find a segment which begins after this one does. 166 */ 167 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 168 q = (struct tcpiphdr *)q->ti_next) 169 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 170 break; 171 172 /* 173 * If there is a preceding segment, it may provide some of 174 * our data already. If so, drop the data from the incoming 175 * segment. If it provides all of our data, drop us. 176 */ 177 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 178 register int i; 179 q = (struct tcpiphdr *)q->ti_prev; 180 /* conversion to int (in i) handles seq wraparound */ 181 i = q->ti_seq + q->ti_len - ti->ti_seq; 182 if (i > 0) { 183 if (i >= ti->ti_len) { 184 tcpstat.tcps_rcvduppack++; 185 tcpstat.tcps_rcvdupbyte += ti->ti_len; 186 m_freem(m); 187 /* 188 * Try to present any queued data 189 * at the left window edge to the user. 190 * This is needed after the 3-WHS 191 * completes. 192 */ 193 goto present; /* ??? */ 194 } 195 m_adj(m, i); 196 ti->ti_len -= i; 197 ti->ti_seq += i; 198 } 199 q = (struct tcpiphdr *)(q->ti_next); 200 } 201 tcpstat.tcps_rcvoopack++; 202 tcpstat.tcps_rcvoobyte += ti->ti_len; 203 REASS_MBUF(ti) = m; /* XXX */ 204 205 /* 206 * While we overlap succeeding segments trim them or, 207 * if they are completely covered, dequeue them. 208 */ 209 while (q != (struct tcpiphdr *)tp) { 210 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 211 if (i <= 0) 212 break; 213 if (i < q->ti_len) { 214 q->ti_seq += i; 215 q->ti_len -= i; 216 m_adj(REASS_MBUF(q), i); 217 break; 218 } 219 q = (struct tcpiphdr *)q->ti_next; 220 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 221 remque(q->ti_prev); 222 m_freem(m); 223 } 224 225 /* 226 * Stick new segment in its place. 227 */ 228 insque(ti, q->ti_prev); 229 230 present: 231 /* 232 * Present data to user, advancing rcv_nxt through 233 * completed sequence space. 234 */ 235 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 236 return (0); 237 ti = tp->seg_next; 238 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 239 return (0); 240 do { 241 tp->rcv_nxt += ti->ti_len; 242 flags = ti->ti_flags & TH_FIN; 243 remque(ti); 244 m = REASS_MBUF(ti); 245 ti = (struct tcpiphdr *)ti->ti_next; 246 if (so->so_state & SS_CANTRCVMORE) 247 m_freem(m); 248 else 249 sbappend(&so->so_rcv, m); 250 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 251 sorwakeup(so); 252 return (flags); 253 } 254 255 /* 256 * TCP input routine, follows pages 65-76 of the 257 * protocol specification dated September, 1981 very closely. 258 */ 259 void 260 tcp_input(m, iphlen) 261 register struct mbuf *m; 262 int iphlen; 263 { 264 register struct tcpiphdr *ti; 265 register struct inpcb *inp; 266 u_char *optp = NULL; 267 int optlen = 0; 268 int len, tlen, off; 269 register struct tcpcb *tp = 0; 270 register int tiflags; 271 struct socket *so = 0; 272 int todrop, acked, ourfinisacked, needoutput = 0; 273 struct in_addr laddr; 274 int dropsocket = 0; 275 int iss = 0; 276 u_long tiwin; 277 struct tcpopt to; /* options in this segment */ 278 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 279 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 280 #ifdef TCPDEBUG 281 short ostate = 0; 282 #endif 283 284 bzero((char *)&to, sizeof(to)); 285 286 tcpstat.tcps_rcvtotal++; 287 /* 288 * Get IP and TCP header together in first mbuf. 289 * Note: IP leaves IP header in first mbuf. 290 */ 291 ti = mtod(m, struct tcpiphdr *); 292 if (iphlen > sizeof (struct ip)) 293 ip_stripoptions(m, (struct mbuf *)0); 294 if (m->m_len < sizeof (struct tcpiphdr)) { 295 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 296 tcpstat.tcps_rcvshort++; 297 return; 298 } 299 ti = mtod(m, struct tcpiphdr *); 300 } 301 302 /* 303 * Checksum extended TCP header and data. 304 */ 305 tlen = ((struct ip *)ti)->ip_len; 306 len = sizeof (struct ip) + tlen; 307 ti->ti_next = ti->ti_prev = 0; 308 ti->ti_x1 = 0; 309 ti->ti_len = (u_short)tlen; 310 HTONS(ti->ti_len); 311 ti->ti_sum = in_cksum(m, len); 312 if (ti->ti_sum) { 313 tcpstat.tcps_rcvbadsum++; 314 goto drop; 315 } 316 #endif /* TUBA_INCLUDE */ 317 318 /* 319 * Check that TCP offset makes sense, 320 * pull out TCP options and adjust length. XXX 321 */ 322 off = ti->ti_off << 2; 323 if (off < sizeof (struct tcphdr) || off > tlen) { 324 tcpstat.tcps_rcvbadoff++; 325 goto drop; 326 } 327 tlen -= off; 328 ti->ti_len = tlen; 329 if (off > sizeof (struct tcphdr)) { 330 if (m->m_len < sizeof(struct ip) + off) { 331 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 332 tcpstat.tcps_rcvshort++; 333 return; 334 } 335 ti = mtod(m, struct tcpiphdr *); 336 } 337 optlen = off - sizeof (struct tcphdr); 338 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 339 } 340 tiflags = ti->ti_flags; 341 342 /* 343 * Convert TCP protocol specific fields to host format. 344 */ 345 NTOHL(ti->ti_seq); 346 NTOHL(ti->ti_ack); 347 NTOHS(ti->ti_win); 348 NTOHS(ti->ti_urp); 349 350 /* 351 * Drop TCP, IP headers and TCP options. 352 */ 353 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 354 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 355 356 /* 357 * Locate pcb for segment. 358 */ 359 findpcb: 360 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 361 ti->ti_dst, ti->ti_dport, 1); 362 363 /* 364 * If the state is CLOSED (i.e., TCB does not exist) then 365 * all data in the incoming segment is discarded. 366 * If the TCB exists but is in CLOSED state, it is embryonic, 367 * but should either do a listen or a connect soon. 368 */ 369 if (inp == NULL) { 370 if (log_in_vain && tiflags & TH_SYN) { 371 char buf[4*sizeof "123"]; 372 373 strcpy(buf, inet_ntoa(ti->ti_dst)); 374 log(LOG_INFO, "Connection attempt to TCP %s:%d" 375 " from %s:%d\n", 376 buf, ntohs(ti->ti_dport), 377 inet_ntoa(ti->ti_src), ntohs(ti->ti_sport)); 378 } 379 goto dropwithreset; 380 } 381 tp = intotcpcb(inp); 382 if (tp == 0) 383 goto dropwithreset; 384 if (tp->t_state == TCPS_CLOSED) 385 goto drop; 386 387 /* Unscale the window into a 32-bit value. */ 388 if ((tiflags & TH_SYN) == 0) 389 tiwin = ti->ti_win << tp->snd_scale; 390 else 391 tiwin = ti->ti_win; 392 393 so = inp->inp_socket; 394 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 395 #ifdef TCPDEBUG 396 if (so->so_options & SO_DEBUG) { 397 ostate = tp->t_state; 398 tcp_saveti = *ti; 399 } 400 #endif 401 if (so->so_options & SO_ACCEPTCONN) { 402 register struct tcpcb *tp0 = tp; 403 struct socket *so2; 404 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 405 /* 406 * Note: dropwithreset makes sure we don't 407 * send a RST in response to a RST. 408 */ 409 if (tiflags & TH_ACK) { 410 tcpstat.tcps_badsyn++; 411 goto dropwithreset; 412 } 413 goto drop; 414 } 415 so2 = sonewconn(so, 0); 416 if (so2 == 0) { 417 tcpstat.tcps_listendrop++; 418 so2 = sodropablereq(so); 419 if (so2) { 420 tcp_drop(sototcpcb(so2), ETIMEDOUT); 421 so2 = sonewconn(so, 0); 422 } 423 if (!so2) 424 goto drop; 425 } 426 so = so2; 427 /* 428 * This is ugly, but .... 429 * 430 * Mark socket as temporary until we're 431 * committed to keeping it. The code at 432 * ``drop'' and ``dropwithreset'' check the 433 * flag dropsocket to see if the temporary 434 * socket created here should be discarded. 435 * We mark the socket as discardable until 436 * we're committed to it below in TCPS_LISTEN. 437 */ 438 dropsocket++; 439 inp = (struct inpcb *)so->so_pcb; 440 inp->inp_laddr = ti->ti_dst; 441 inp->inp_lport = ti->ti_dport; 442 in_pcbrehash(inp); 443 #if BSD>=43 444 inp->inp_options = ip_srcroute(); 445 #endif 446 tp = intotcpcb(inp); 447 tp->t_state = TCPS_LISTEN; 448 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 449 450 /* Compute proper scaling value from buffer space */ 451 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 452 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 453 tp->request_r_scale++; 454 } 455 } 456 457 /* 458 * Segment received on connection. 459 * Reset idle time and keep-alive timer. 460 */ 461 tp->t_idle = 0; 462 if (TCPS_HAVEESTABLISHED(tp->t_state)) 463 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 464 465 /* 466 * Process options if not in LISTEN state, 467 * else do it below (after getting remote address). 468 */ 469 if (tp->t_state != TCPS_LISTEN) 470 tcp_dooptions(tp, optp, optlen, ti, &to); 471 472 /* 473 * Header prediction: check for the two common cases 474 * of a uni-directional data xfer. If the packet has 475 * no control flags, is in-sequence, the window didn't 476 * change and we're not retransmitting, it's a 477 * candidate. If the length is zero and the ack moved 478 * forward, we're the sender side of the xfer. Just 479 * free the data acked & wake any higher level process 480 * that was blocked waiting for space. If the length 481 * is non-zero and the ack didn't move, we're the 482 * receiver side. If we're getting packets in-order 483 * (the reassembly queue is empty), add the data to 484 * the socket buffer and note that we need a delayed ack. 485 * Make sure that the hidden state-flags are also off. 486 * Since we check for TCPS_ESTABLISHED above, it can only 487 * be TH_NEEDSYN. 488 */ 489 if (tp->t_state == TCPS_ESTABLISHED && 490 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 491 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 492 ((to.to_flag & TOF_TS) == 0 || 493 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 494 /* 495 * Using the CC option is compulsory if once started: 496 * the segment is OK if no T/TCP was negotiated or 497 * if the segment has a CC option equal to CCrecv 498 */ 499 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 500 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 501 ti->ti_seq == tp->rcv_nxt && 502 tiwin && tiwin == tp->snd_wnd && 503 tp->snd_nxt == tp->snd_max) { 504 505 /* 506 * If last ACK falls within this segment's sequence numbers, 507 * record the timestamp. 508 * NOTE that the test is modified according to the latest 509 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 510 */ 511 if ((to.to_flag & TOF_TS) != 0 && 512 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 513 tp->ts_recent_age = tcp_now; 514 tp->ts_recent = to.to_tsval; 515 } 516 517 if (ti->ti_len == 0) { 518 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 519 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 520 tp->snd_cwnd >= tp->snd_wnd && 521 tp->t_dupacks < tcprexmtthresh) { 522 /* 523 * this is a pure ack for outstanding data. 524 */ 525 ++tcpstat.tcps_predack; 526 if ((to.to_flag & TOF_TS) != 0) 527 tcp_xmit_timer(tp, 528 tcp_now - to.to_tsecr + 1); 529 else if (tp->t_rtt && 530 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 531 tcp_xmit_timer(tp, tp->t_rtt); 532 acked = ti->ti_ack - tp->snd_una; 533 tcpstat.tcps_rcvackpack++; 534 tcpstat.tcps_rcvackbyte += acked; 535 sbdrop(&so->so_snd, acked); 536 tp->snd_una = ti->ti_ack; 537 m_freem(m); 538 539 /* 540 * If all outstanding data are acked, stop 541 * retransmit timer, otherwise restart timer 542 * using current (possibly backed-off) value. 543 * If process is waiting for space, 544 * wakeup/selwakeup/signal. If data 545 * are ready to send, let tcp_output 546 * decide between more output or persist. 547 */ 548 if (tp->snd_una == tp->snd_max) 549 tp->t_timer[TCPT_REXMT] = 0; 550 else if (tp->t_timer[TCPT_PERSIST] == 0) 551 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 552 553 if (so->so_snd.sb_flags & SB_NOTIFY) 554 sowwakeup(so); 555 if (so->so_snd.sb_cc) 556 (void) tcp_output(tp); 557 return; 558 } 559 } else if (ti->ti_ack == tp->snd_una && 560 tp->seg_next == (struct tcpiphdr *)tp && 561 ti->ti_len <= sbspace(&so->so_rcv)) { 562 /* 563 * this is a pure, in-sequence data packet 564 * with nothing on the reassembly queue and 565 * we have enough buffer space to take it. 566 */ 567 ++tcpstat.tcps_preddat; 568 tp->rcv_nxt += ti->ti_len; 569 tcpstat.tcps_rcvpack++; 570 tcpstat.tcps_rcvbyte += ti->ti_len; 571 /* 572 * Add data to socket buffer. 573 */ 574 sbappend(&so->so_rcv, m); 575 sorwakeup(so); 576 #ifdef TCP_ACK_HACK 577 /* 578 * If this is a short packet, then ACK now - with Nagel 579 * congestion avoidance sender won't send more until 580 * he gets an ACK. 581 */ 582 if (tiflags & TH_PUSH) { 583 tp->t_flags |= TF_ACKNOW; 584 tcp_output(tp); 585 } else { 586 tp->t_flags |= TF_DELACK; 587 } 588 #else 589 tp->t_flags |= TF_DELACK; 590 #endif 591 return; 592 } 593 } 594 595 /* 596 * Calculate amount of space in receive window, 597 * and then do TCP input processing. 598 * Receive window is amount of space in rcv queue, 599 * but not less than advertised window. 600 */ 601 { int win; 602 603 win = sbspace(&so->so_rcv); 604 if (win < 0) 605 win = 0; 606 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 607 } 608 609 switch (tp->t_state) { 610 611 /* 612 * If the state is LISTEN then ignore segment if it contains an RST. 613 * If the segment contains an ACK then it is bad and send a RST. 614 * If it does not contain a SYN then it is not interesting; drop it. 615 * Don't bother responding if the destination was a broadcast. 616 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 617 * tp->iss, and send a segment: 618 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 619 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 620 * Fill in remote peer address fields if not previously specified. 621 * Enter SYN_RECEIVED state, and process any other fields of this 622 * segment in this state. 623 */ 624 case TCPS_LISTEN: { 625 struct mbuf *am; 626 register struct sockaddr_in *sin; 627 628 if (tiflags & TH_RST) 629 goto drop; 630 if (tiflags & TH_ACK) 631 goto dropwithreset; 632 if ((tiflags & TH_SYN) == 0) 633 goto drop; 634 /* 635 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 636 * in_broadcast() should never return true on a received 637 * packet with M_BCAST not set. 638 */ 639 if (m->m_flags & (M_BCAST|M_MCAST) || 640 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 641 goto drop; 642 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 643 if (am == NULL) 644 goto drop; 645 am->m_len = sizeof (struct sockaddr_in); 646 sin = mtod(am, struct sockaddr_in *); 647 sin->sin_family = AF_INET; 648 sin->sin_len = sizeof(*sin); 649 sin->sin_addr = ti->ti_src; 650 sin->sin_port = ti->ti_sport; 651 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 652 laddr = inp->inp_laddr; 653 if (inp->inp_laddr.s_addr == INADDR_ANY) 654 inp->inp_laddr = ti->ti_dst; 655 if (in_pcbconnect(inp, am)) { 656 inp->inp_laddr = laddr; 657 (void) m_free(am); 658 goto drop; 659 } 660 (void) m_free(am); 661 tp->t_template = tcp_template(tp); 662 if (tp->t_template == 0) { 663 tp = tcp_drop(tp, ENOBUFS); 664 dropsocket = 0; /* socket is already gone */ 665 goto drop; 666 } 667 if ((taop = tcp_gettaocache(inp)) == NULL) { 668 taop = &tao_noncached; 669 bzero(taop, sizeof(*taop)); 670 } 671 tcp_dooptions(tp, optp, optlen, ti, &to); 672 if (iss) 673 tp->iss = iss; 674 else 675 tp->iss = tcp_iss; 676 tcp_iss += TCP_ISSINCR/4; 677 tp->irs = ti->ti_seq; 678 tcp_sendseqinit(tp); 679 tcp_rcvseqinit(tp); 680 /* 681 * Initialization of the tcpcb for transaction; 682 * set SND.WND = SEG.WND, 683 * initialize CCsend and CCrecv. 684 */ 685 tp->snd_wnd = tiwin; /* initial send-window */ 686 tp->cc_send = CC_INC(tcp_ccgen); 687 tp->cc_recv = to.to_cc; 688 /* 689 * Perform TAO test on incoming CC (SEG.CC) option, if any. 690 * - compare SEG.CC against cached CC from the same host, 691 * if any. 692 * - if SEG.CC > chached value, SYN must be new and is accepted 693 * immediately: save new CC in the cache, mark the socket 694 * connected, enter ESTABLISHED state, turn on flag to 695 * send a SYN in the next segment. 696 * A virtual advertised window is set in rcv_adv to 697 * initialize SWS prevention. Then enter normal segment 698 * processing: drop SYN, process data and FIN. 699 * - otherwise do a normal 3-way handshake. 700 */ 701 if ((to.to_flag & TOF_CC) != 0) { 702 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 703 taop->tao_cc = to.to_cc; 704 tp->t_state = TCPS_ESTABLISHED; 705 706 /* 707 * If there is a FIN, or if there is data and the 708 * connection is local, then delay SYN,ACK(SYN) in 709 * the hope of piggy-backing it on a response 710 * segment. Otherwise must send ACK now in case 711 * the other side is slow starting. 712 */ 713 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 714 in_localaddr(inp->inp_faddr))) 715 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 716 else 717 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 718 719 /* 720 * Limit the `virtual advertised window' to TCP_MAXWIN 721 * here. Even if we requested window scaling, it will 722 * become effective only later when our SYN is acked. 723 */ 724 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 725 tcpstat.tcps_connects++; 726 soisconnected(so); 727 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 728 dropsocket = 0; /* committed to socket */ 729 tcpstat.tcps_accepts++; 730 goto trimthenstep6; 731 } 732 /* else do standard 3-way handshake */ 733 } else { 734 /* 735 * No CC option, but maybe CC.NEW: 736 * invalidate cached value. 737 */ 738 taop->tao_cc = 0; 739 } 740 /* 741 * TAO test failed or there was no CC option, 742 * do a standard 3-way handshake. 743 */ 744 tp->t_flags |= TF_ACKNOW; 745 tp->t_state = TCPS_SYN_RECEIVED; 746 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 747 dropsocket = 0; /* committed to socket */ 748 tcpstat.tcps_accepts++; 749 goto trimthenstep6; 750 } 751 752 /* 753 * If the state is SYN_RECEIVED: 754 * do just the ack and RST checks from SYN_SENT state. 755 * If the state is SYN_SENT: 756 * if seg contains an ACK, but not for our SYN, drop the input. 757 * if seg contains a RST, then drop the connection. 758 * if seg does not contain SYN, then drop it. 759 * Otherwise this is an acceptable SYN segment 760 * initialize tp->rcv_nxt and tp->irs 761 * if seg contains ack then advance tp->snd_una 762 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 763 * arrange for segment to be acked (eventually) 764 * continue processing rest of data/controls, beginning with URG 765 */ 766 case TCPS_SYN_RECEIVED: 767 case TCPS_SYN_SENT: 768 if ((taop = tcp_gettaocache(inp)) == NULL) { 769 taop = &tao_noncached; 770 bzero(taop, sizeof(*taop)); 771 } 772 773 if ((tiflags & TH_ACK) && 774 (SEQ_LEQ(ti->ti_ack, tp->iss) || 775 SEQ_GT(ti->ti_ack, tp->snd_max))) { 776 /* 777 * If we have a cached CCsent for the remote host, 778 * hence we haven't just crashed and restarted, 779 * do not send a RST. This may be a retransmission 780 * from the other side after our earlier ACK was lost. 781 * Our new SYN, when it arrives, will serve as the 782 * needed ACK. 783 */ 784 if (taop->tao_ccsent != 0) 785 goto drop; 786 else 787 goto dropwithreset; 788 } 789 if (tiflags & TH_RST) { 790 if (tiflags & TH_ACK) 791 tp = tcp_drop(tp, ECONNREFUSED); 792 goto drop; 793 } 794 if (tp->t_state == TCPS_SYN_RECEIVED) 795 break; 796 if ((tiflags & TH_SYN) == 0) 797 goto drop; 798 tp->snd_wnd = ti->ti_win; /* initial send window */ 799 tp->cc_recv = to.to_cc; /* foreign CC */ 800 801 tp->irs = ti->ti_seq; 802 tcp_rcvseqinit(tp); 803 if (tiflags & TH_ACK) { 804 /* 805 * Our SYN was acked. If segment contains CC.ECHO 806 * option, check it to make sure this segment really 807 * matches our SYN. If not, just drop it as old 808 * duplicate, but send an RST if we're still playing 809 * by the old rules. 810 */ 811 if ((to.to_flag & TOF_CCECHO) && 812 tp->cc_send != to.to_ccecho) { 813 if (taop->tao_ccsent != 0) 814 goto drop; 815 else 816 goto dropwithreset; 817 } 818 tcpstat.tcps_connects++; 819 soisconnected(so); 820 /* Do window scaling on this connection? */ 821 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 822 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 823 tp->snd_scale = tp->requested_s_scale; 824 tp->rcv_scale = tp->request_r_scale; 825 } 826 /* Segment is acceptable, update cache if undefined. */ 827 if (taop->tao_ccsent == 0) 828 taop->tao_ccsent = to.to_ccecho; 829 830 tp->rcv_adv += tp->rcv_wnd; 831 tp->snd_una++; /* SYN is acked */ 832 /* 833 * If there's data, delay ACK; if there's also a FIN 834 * ACKNOW will be turned on later. 835 */ 836 if (ti->ti_len != 0) 837 tp->t_flags |= TF_DELACK; 838 else 839 tp->t_flags |= TF_ACKNOW; 840 /* 841 * Received <SYN,ACK> in SYN_SENT[*] state. 842 * Transitions: 843 * SYN_SENT --> ESTABLISHED 844 * SYN_SENT* --> FIN_WAIT_1 845 */ 846 if (tp->t_flags & TF_NEEDFIN) { 847 tp->t_state = TCPS_FIN_WAIT_1; 848 tp->t_flags &= ~TF_NEEDFIN; 849 tiflags &= ~TH_SYN; 850 } else { 851 tp->t_state = TCPS_ESTABLISHED; 852 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 853 } 854 } else { 855 /* 856 * Received initial SYN in SYN-SENT[*] state => simul- 857 * taneous open. If segment contains CC option and there is 858 * a cached CC, apply TAO test; if it succeeds, connection is 859 * half-synchronized. Otherwise, do 3-way handshake: 860 * SYN-SENT -> SYN-RECEIVED 861 * SYN-SENT* -> SYN-RECEIVED* 862 * If there was no CC option, clear cached CC value. 863 */ 864 tp->t_flags |= TF_ACKNOW; 865 tp->t_timer[TCPT_REXMT] = 0; 866 if (to.to_flag & TOF_CC) { 867 if (taop->tao_cc != 0 && 868 CC_GT(to.to_cc, taop->tao_cc)) { 869 /* 870 * update cache and make transition: 871 * SYN-SENT -> ESTABLISHED* 872 * SYN-SENT* -> FIN-WAIT-1* 873 */ 874 taop->tao_cc = to.to_cc; 875 if (tp->t_flags & TF_NEEDFIN) { 876 tp->t_state = TCPS_FIN_WAIT_1; 877 tp->t_flags &= ~TF_NEEDFIN; 878 } else { 879 tp->t_state = TCPS_ESTABLISHED; 880 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 881 } 882 tp->t_flags |= TF_NEEDSYN; 883 } else 884 tp->t_state = TCPS_SYN_RECEIVED; 885 } else { 886 /* CC.NEW or no option => invalidate cache */ 887 taop->tao_cc = 0; 888 tp->t_state = TCPS_SYN_RECEIVED; 889 } 890 } 891 892 trimthenstep6: 893 /* 894 * Advance ti->ti_seq to correspond to first data byte. 895 * If data, trim to stay within window, 896 * dropping FIN if necessary. 897 */ 898 ti->ti_seq++; 899 if (ti->ti_len > tp->rcv_wnd) { 900 todrop = ti->ti_len - tp->rcv_wnd; 901 m_adj(m, -todrop); 902 ti->ti_len = tp->rcv_wnd; 903 tiflags &= ~TH_FIN; 904 tcpstat.tcps_rcvpackafterwin++; 905 tcpstat.tcps_rcvbyteafterwin += todrop; 906 } 907 tp->snd_wl1 = ti->ti_seq - 1; 908 tp->rcv_up = ti->ti_seq; 909 /* 910 * Client side of transaction: already sent SYN and data. 911 * If the remote host used T/TCP to validate the SYN, 912 * our data will be ACK'd; if so, enter normal data segment 913 * processing in the middle of step 5, ack processing. 914 * Otherwise, goto step 6. 915 */ 916 if (tiflags & TH_ACK) 917 goto process_ACK; 918 goto step6; 919 /* 920 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 921 * if segment contains a SYN and CC [not CC.NEW] option: 922 * if state == TIME_WAIT and connection duration > MSL, 923 * drop packet and send RST; 924 * 925 * if SEG.CC > CCrecv then is new SYN, and can implicitly 926 * ack the FIN (and data) in retransmission queue. 927 * Complete close and delete TCPCB. Then reprocess 928 * segment, hoping to find new TCPCB in LISTEN state; 929 * 930 * else must be old SYN; drop it. 931 * else do normal processing. 932 */ 933 case TCPS_LAST_ACK: 934 case TCPS_CLOSING: 935 case TCPS_TIME_WAIT: 936 if ((tiflags & TH_SYN) && 937 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 938 if (tp->t_state == TCPS_TIME_WAIT && 939 tp->t_duration > TCPTV_MSL) 940 goto dropwithreset; 941 if (CC_GT(to.to_cc, tp->cc_recv)) { 942 tp = tcp_close(tp); 943 goto findpcb; 944 } 945 else 946 goto drop; 947 } 948 break; /* continue normal processing */ 949 } 950 951 /* 952 * States other than LISTEN or SYN_SENT. 953 * First check timestamp, if present. 954 * Then check the connection count, if present. 955 * Then check that at least some bytes of segment are within 956 * receive window. If segment begins before rcv_nxt, 957 * drop leading data (and SYN); if nothing left, just ack. 958 * 959 * RFC 1323 PAWS: If we have a timestamp reply on this segment 960 * and it's less than ts_recent, drop it. 961 */ 962 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 963 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 964 965 /* Check to see if ts_recent is over 24 days old. */ 966 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 967 /* 968 * Invalidate ts_recent. If this segment updates 969 * ts_recent, the age will be reset later and ts_recent 970 * will get a valid value. If it does not, setting 971 * ts_recent to zero will at least satisfy the 972 * requirement that zero be placed in the timestamp 973 * echo reply when ts_recent isn't valid. The 974 * age isn't reset until we get a valid ts_recent 975 * because we don't want out-of-order segments to be 976 * dropped when ts_recent is old. 977 */ 978 tp->ts_recent = 0; 979 } else { 980 tcpstat.tcps_rcvduppack++; 981 tcpstat.tcps_rcvdupbyte += ti->ti_len; 982 tcpstat.tcps_pawsdrop++; 983 goto dropafterack; 984 } 985 } 986 987 /* 988 * T/TCP mechanism 989 * If T/TCP was negotiated and the segment doesn't have CC, 990 * or if it's CC is wrong then drop the segment. 991 * RST segments do not have to comply with this. 992 */ 993 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 994 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 995 (tiflags & TH_RST) == 0) 996 goto dropafterack; 997 998 todrop = tp->rcv_nxt - ti->ti_seq; 999 if (todrop > 0) { 1000 if (tiflags & TH_SYN) { 1001 tiflags &= ~TH_SYN; 1002 ti->ti_seq++; 1003 if (ti->ti_urp > 1) 1004 ti->ti_urp--; 1005 else 1006 tiflags &= ~TH_URG; 1007 todrop--; 1008 } 1009 /* 1010 * Following if statement from Stevens, vol. 2, p. 960. 1011 */ 1012 if (todrop > ti->ti_len 1013 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1014 /* 1015 * Any valid FIN must be to the left of the window. 1016 * At this point the FIN must be a duplicate or out 1017 * of sequence; drop it. 1018 */ 1019 tiflags &= ~TH_FIN; 1020 1021 /* 1022 * Send an ACK to resynchronize and drop any data. 1023 * But keep on processing for RST or ACK. 1024 */ 1025 tp->t_flags |= TF_ACKNOW; 1026 todrop = ti->ti_len; 1027 tcpstat.tcps_rcvduppack++; 1028 tcpstat.tcps_rcvdupbyte += todrop; 1029 } else { 1030 tcpstat.tcps_rcvpartduppack++; 1031 tcpstat.tcps_rcvpartdupbyte += todrop; 1032 } 1033 m_adj(m, todrop); 1034 ti->ti_seq += todrop; 1035 ti->ti_len -= todrop; 1036 if (ti->ti_urp > todrop) 1037 ti->ti_urp -= todrop; 1038 else { 1039 tiflags &= ~TH_URG; 1040 ti->ti_urp = 0; 1041 } 1042 } 1043 1044 /* 1045 * If new data are received on a connection after the 1046 * user processes are gone, then RST the other end. 1047 */ 1048 if ((so->so_state & SS_NOFDREF) && 1049 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1050 tp = tcp_close(tp); 1051 tcpstat.tcps_rcvafterclose++; 1052 goto dropwithreset; 1053 } 1054 1055 /* 1056 * If segment ends after window, drop trailing data 1057 * (and PUSH and FIN); if nothing left, just ACK. 1058 */ 1059 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1060 if (todrop > 0) { 1061 tcpstat.tcps_rcvpackafterwin++; 1062 if (todrop >= ti->ti_len) { 1063 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1064 /* 1065 * If a new connection request is received 1066 * while in TIME_WAIT, drop the old connection 1067 * and start over if the sequence numbers 1068 * are above the previous ones. 1069 */ 1070 if (tiflags & TH_SYN && 1071 tp->t_state == TCPS_TIME_WAIT && 1072 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1073 iss = tp->rcv_nxt + TCP_ISSINCR; 1074 tp = tcp_close(tp); 1075 goto findpcb; 1076 } 1077 /* 1078 * If window is closed can only take segments at 1079 * window edge, and have to drop data and PUSH from 1080 * incoming segments. Continue processing, but 1081 * remember to ack. Otherwise, drop segment 1082 * and ack. 1083 */ 1084 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1085 tp->t_flags |= TF_ACKNOW; 1086 tcpstat.tcps_rcvwinprobe++; 1087 } else 1088 goto dropafterack; 1089 } else 1090 tcpstat.tcps_rcvbyteafterwin += todrop; 1091 m_adj(m, -todrop); 1092 ti->ti_len -= todrop; 1093 tiflags &= ~(TH_PUSH|TH_FIN); 1094 } 1095 1096 /* 1097 * If last ACK falls within this segment's sequence numbers, 1098 * record its timestamp. 1099 * NOTE that the test is modified according to the latest 1100 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1101 */ 1102 if ((to.to_flag & TOF_TS) != 0 && 1103 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1104 tp->ts_recent_age = tcp_now; 1105 tp->ts_recent = to.to_tsval; 1106 } 1107 1108 /* 1109 * If the RST bit is set examine the state: 1110 * SYN_RECEIVED STATE: 1111 * If passive open, return to LISTEN state. 1112 * If active open, inform user that connection was refused. 1113 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1114 * Inform user that connection was reset, and close tcb. 1115 * CLOSING, LAST_ACK, TIME_WAIT STATES 1116 * Close the tcb. 1117 */ 1118 if (tiflags&TH_RST) switch (tp->t_state) { 1119 1120 case TCPS_SYN_RECEIVED: 1121 so->so_error = ECONNREFUSED; 1122 goto close; 1123 1124 case TCPS_ESTABLISHED: 1125 case TCPS_FIN_WAIT_1: 1126 case TCPS_FIN_WAIT_2: 1127 case TCPS_CLOSE_WAIT: 1128 so->so_error = ECONNRESET; 1129 close: 1130 tp->t_state = TCPS_CLOSED; 1131 tcpstat.tcps_drops++; 1132 tp = tcp_close(tp); 1133 goto drop; 1134 1135 case TCPS_CLOSING: 1136 case TCPS_LAST_ACK: 1137 case TCPS_TIME_WAIT: 1138 tp = tcp_close(tp); 1139 goto drop; 1140 } 1141 1142 /* 1143 * If a SYN is in the window, then this is an 1144 * error and we send an RST and drop the connection. 1145 */ 1146 if (tiflags & TH_SYN) { 1147 tp = tcp_drop(tp, ECONNRESET); 1148 goto dropwithreset; 1149 } 1150 1151 /* 1152 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1153 * flag is on (half-synchronized state), then queue data for 1154 * later processing; else drop segment and return. 1155 */ 1156 if ((tiflags & TH_ACK) == 0) { 1157 if (tp->t_state == TCPS_SYN_RECEIVED || 1158 (tp->t_flags & TF_NEEDSYN)) 1159 goto step6; 1160 else 1161 goto drop; 1162 } 1163 1164 /* 1165 * Ack processing. 1166 */ 1167 switch (tp->t_state) { 1168 1169 /* 1170 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1171 * ESTABLISHED state and continue processing, otherwise 1172 * send an RST. 1173 */ 1174 case TCPS_SYN_RECEIVED: 1175 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1176 SEQ_GT(ti->ti_ack, tp->snd_max)) 1177 goto dropwithreset; 1178 1179 tcpstat.tcps_connects++; 1180 soisconnected(so); 1181 /* Do window scaling? */ 1182 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1183 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1184 tp->snd_scale = tp->requested_s_scale; 1185 tp->rcv_scale = tp->request_r_scale; 1186 } 1187 /* 1188 * Upon successful completion of 3-way handshake, 1189 * update cache.CC if it was undefined, pass any queued 1190 * data to the user, and advance state appropriately. 1191 */ 1192 if ((taop = tcp_gettaocache(inp)) != NULL && 1193 taop->tao_cc == 0) 1194 taop->tao_cc = tp->cc_recv; 1195 1196 /* 1197 * Make transitions: 1198 * SYN-RECEIVED -> ESTABLISHED 1199 * SYN-RECEIVED* -> FIN-WAIT-1 1200 */ 1201 if (tp->t_flags & TF_NEEDFIN) { 1202 tp->t_state = TCPS_FIN_WAIT_1; 1203 tp->t_flags &= ~TF_NEEDFIN; 1204 } else { 1205 tp->t_state = TCPS_ESTABLISHED; 1206 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1207 } 1208 /* 1209 * If segment contains data or ACK, will call tcp_reass() 1210 * later; if not, do so now to pass queued data to user. 1211 */ 1212 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1213 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1214 (struct mbuf *)0); 1215 tp->snd_wl1 = ti->ti_seq - 1; 1216 /* fall into ... */ 1217 1218 /* 1219 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1220 * ACKs. If the ack is in the range 1221 * tp->snd_una < ti->ti_ack <= tp->snd_max 1222 * then advance tp->snd_una to ti->ti_ack and drop 1223 * data from the retransmission queue. If this ACK reflects 1224 * more up to date window information we update our window information. 1225 */ 1226 case TCPS_ESTABLISHED: 1227 case TCPS_FIN_WAIT_1: 1228 case TCPS_FIN_WAIT_2: 1229 case TCPS_CLOSE_WAIT: 1230 case TCPS_CLOSING: 1231 case TCPS_LAST_ACK: 1232 case TCPS_TIME_WAIT: 1233 1234 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1235 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1236 tcpstat.tcps_rcvdupack++; 1237 /* 1238 * If we have outstanding data (other than 1239 * a window probe), this is a completely 1240 * duplicate ack (ie, window info didn't 1241 * change), the ack is the biggest we've 1242 * seen and we've seen exactly our rexmt 1243 * threshhold of them, assume a packet 1244 * has been dropped and retransmit it. 1245 * Kludge snd_nxt & the congestion 1246 * window so we send only this one 1247 * packet. 1248 * 1249 * We know we're losing at the current 1250 * window size so do congestion avoidance 1251 * (set ssthresh to half the current window 1252 * and pull our congestion window back to 1253 * the new ssthresh). 1254 * 1255 * Dup acks mean that packets have left the 1256 * network (they're now cached at the receiver) 1257 * so bump cwnd by the amount in the receiver 1258 * to keep a constant cwnd packets in the 1259 * network. 1260 */ 1261 if (tp->t_timer[TCPT_REXMT] == 0 || 1262 ti->ti_ack != tp->snd_una) 1263 tp->t_dupacks = 0; 1264 else if (++tp->t_dupacks == tcprexmtthresh) { 1265 tcp_seq onxt = tp->snd_nxt; 1266 u_int win = 1267 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1268 tp->t_maxseg; 1269 1270 if (win < 2) 1271 win = 2; 1272 tp->snd_ssthresh = win * tp->t_maxseg; 1273 tp->t_timer[TCPT_REXMT] = 0; 1274 tp->t_rtt = 0; 1275 tp->snd_nxt = ti->ti_ack; 1276 tp->snd_cwnd = tp->t_maxseg; 1277 (void) tcp_output(tp); 1278 tp->snd_cwnd = tp->snd_ssthresh + 1279 tp->t_maxseg * tp->t_dupacks; 1280 if (SEQ_GT(onxt, tp->snd_nxt)) 1281 tp->snd_nxt = onxt; 1282 goto drop; 1283 } else if (tp->t_dupacks > tcprexmtthresh) { 1284 tp->snd_cwnd += tp->t_maxseg; 1285 (void) tcp_output(tp); 1286 goto drop; 1287 } 1288 } else 1289 tp->t_dupacks = 0; 1290 break; 1291 } 1292 /* 1293 * If the congestion window was inflated to account 1294 * for the other side's cached packets, retract it. 1295 */ 1296 if (tp->t_dupacks >= tcprexmtthresh && 1297 tp->snd_cwnd > tp->snd_ssthresh) 1298 tp->snd_cwnd = tp->snd_ssthresh; 1299 tp->t_dupacks = 0; 1300 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1301 tcpstat.tcps_rcvacktoomuch++; 1302 goto dropafterack; 1303 } 1304 /* 1305 * If we reach this point, ACK is not a duplicate, 1306 * i.e., it ACKs something we sent. 1307 */ 1308 if (tp->t_flags & TF_NEEDSYN) { 1309 /* 1310 * T/TCP: Connection was half-synchronized, and our 1311 * SYN has been ACK'd (so connection is now fully 1312 * synchronized). Go to non-starred state, 1313 * increment snd_una for ACK of SYN, and check if 1314 * we can do window scaling. 1315 */ 1316 tp->t_flags &= ~TF_NEEDSYN; 1317 tp->snd_una++; 1318 /* Do window scaling? */ 1319 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1320 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1321 tp->snd_scale = tp->requested_s_scale; 1322 tp->rcv_scale = tp->request_r_scale; 1323 } 1324 } 1325 1326 process_ACK: 1327 acked = ti->ti_ack - tp->snd_una; 1328 tcpstat.tcps_rcvackpack++; 1329 tcpstat.tcps_rcvackbyte += acked; 1330 1331 /* 1332 * If we have a timestamp reply, update smoothed 1333 * round trip time. If no timestamp is present but 1334 * transmit timer is running and timed sequence 1335 * number was acked, update smoothed round trip time. 1336 * Since we now have an rtt measurement, cancel the 1337 * timer backoff (cf., Phil Karn's retransmit alg.). 1338 * Recompute the initial retransmit timer. 1339 */ 1340 if (to.to_flag & TOF_TS) 1341 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1342 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1343 tcp_xmit_timer(tp,tp->t_rtt); 1344 1345 /* 1346 * If all outstanding data is acked, stop retransmit 1347 * timer and remember to restart (more output or persist). 1348 * If there is more data to be acked, restart retransmit 1349 * timer, using current (possibly backed-off) value. 1350 */ 1351 if (ti->ti_ack == tp->snd_max) { 1352 tp->t_timer[TCPT_REXMT] = 0; 1353 needoutput = 1; 1354 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1355 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1356 1357 /* 1358 * If no data (only SYN) was ACK'd, 1359 * skip rest of ACK processing. 1360 */ 1361 if (acked == 0) 1362 goto step6; 1363 1364 /* 1365 * When new data is acked, open the congestion window. 1366 * If the window gives us less than ssthresh packets 1367 * in flight, open exponentially (maxseg per packet). 1368 * Otherwise open linearly: maxseg per window 1369 * (maxseg^2 / cwnd per packet). 1370 */ 1371 { 1372 register u_int cw = tp->snd_cwnd; 1373 register u_int incr = tp->t_maxseg; 1374 1375 if (cw > tp->snd_ssthresh) 1376 incr = incr * incr / cw; 1377 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1378 } 1379 if (acked > so->so_snd.sb_cc) { 1380 tp->snd_wnd -= so->so_snd.sb_cc; 1381 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1382 ourfinisacked = 1; 1383 } else { 1384 sbdrop(&so->so_snd, acked); 1385 tp->snd_wnd -= acked; 1386 ourfinisacked = 0; 1387 } 1388 if (so->so_snd.sb_flags & SB_NOTIFY) 1389 sowwakeup(so); 1390 tp->snd_una = ti->ti_ack; 1391 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1392 tp->snd_nxt = tp->snd_una; 1393 1394 switch (tp->t_state) { 1395 1396 /* 1397 * In FIN_WAIT_1 STATE in addition to the processing 1398 * for the ESTABLISHED state if our FIN is now acknowledged 1399 * then enter FIN_WAIT_2. 1400 */ 1401 case TCPS_FIN_WAIT_1: 1402 if (ourfinisacked) { 1403 /* 1404 * If we can't receive any more 1405 * data, then closing user can proceed. 1406 * Starting the timer is contrary to the 1407 * specification, but if we don't get a FIN 1408 * we'll hang forever. 1409 */ 1410 if (so->so_state & SS_CANTRCVMORE) { 1411 soisdisconnected(so); 1412 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1413 } 1414 tp->t_state = TCPS_FIN_WAIT_2; 1415 } 1416 break; 1417 1418 /* 1419 * In CLOSING STATE in addition to the processing for 1420 * the ESTABLISHED state if the ACK acknowledges our FIN 1421 * then enter the TIME-WAIT state, otherwise ignore 1422 * the segment. 1423 */ 1424 case TCPS_CLOSING: 1425 if (ourfinisacked) { 1426 tp->t_state = TCPS_TIME_WAIT; 1427 tcp_canceltimers(tp); 1428 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1429 if (tp->cc_recv != 0 && 1430 tp->t_duration < TCPTV_MSL) 1431 tp->t_timer[TCPT_2MSL] = 1432 tp->t_rxtcur * TCPTV_TWTRUNC; 1433 else 1434 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1435 soisdisconnected(so); 1436 } 1437 break; 1438 1439 /* 1440 * In LAST_ACK, we may still be waiting for data to drain 1441 * and/or to be acked, as well as for the ack of our FIN. 1442 * If our FIN is now acknowledged, delete the TCB, 1443 * enter the closed state and return. 1444 */ 1445 case TCPS_LAST_ACK: 1446 if (ourfinisacked) { 1447 tp = tcp_close(tp); 1448 goto drop; 1449 } 1450 break; 1451 1452 /* 1453 * In TIME_WAIT state the only thing that should arrive 1454 * is a retransmission of the remote FIN. Acknowledge 1455 * it and restart the finack timer. 1456 */ 1457 case TCPS_TIME_WAIT: 1458 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1459 goto dropafterack; 1460 } 1461 } 1462 1463 step6: 1464 /* 1465 * Update window information. 1466 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1467 */ 1468 if ((tiflags & TH_ACK) && 1469 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1470 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1471 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1472 /* keep track of pure window updates */ 1473 if (ti->ti_len == 0 && 1474 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1475 tcpstat.tcps_rcvwinupd++; 1476 tp->snd_wnd = tiwin; 1477 tp->snd_wl1 = ti->ti_seq; 1478 tp->snd_wl2 = ti->ti_ack; 1479 if (tp->snd_wnd > tp->max_sndwnd) 1480 tp->max_sndwnd = tp->snd_wnd; 1481 needoutput = 1; 1482 } 1483 1484 /* 1485 * Process segments with URG. 1486 */ 1487 if ((tiflags & TH_URG) && ti->ti_urp && 1488 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1489 /* 1490 * This is a kludge, but if we receive and accept 1491 * random urgent pointers, we'll crash in 1492 * soreceive. It's hard to imagine someone 1493 * actually wanting to send this much urgent data. 1494 */ 1495 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1496 ti->ti_urp = 0; /* XXX */ 1497 tiflags &= ~TH_URG; /* XXX */ 1498 goto dodata; /* XXX */ 1499 } 1500 /* 1501 * If this segment advances the known urgent pointer, 1502 * then mark the data stream. This should not happen 1503 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1504 * a FIN has been received from the remote side. 1505 * In these states we ignore the URG. 1506 * 1507 * According to RFC961 (Assigned Protocols), 1508 * the urgent pointer points to the last octet 1509 * of urgent data. We continue, however, 1510 * to consider it to indicate the first octet 1511 * of data past the urgent section as the original 1512 * spec states (in one of two places). 1513 */ 1514 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1515 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1516 so->so_oobmark = so->so_rcv.sb_cc + 1517 (tp->rcv_up - tp->rcv_nxt) - 1; 1518 if (so->so_oobmark == 0) 1519 so->so_state |= SS_RCVATMARK; 1520 sohasoutofband(so); 1521 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1522 } 1523 /* 1524 * Remove out of band data so doesn't get presented to user. 1525 * This can happen independent of advancing the URG pointer, 1526 * but if two URG's are pending at once, some out-of-band 1527 * data may creep in... ick. 1528 */ 1529 if (ti->ti_urp <= (u_long)ti->ti_len 1530 #ifdef SO_OOBINLINE 1531 && (so->so_options & SO_OOBINLINE) == 0 1532 #endif 1533 ) 1534 tcp_pulloutofband(so, ti, m); 1535 } else 1536 /* 1537 * If no out of band data is expected, 1538 * pull receive urgent pointer along 1539 * with the receive window. 1540 */ 1541 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1542 tp->rcv_up = tp->rcv_nxt; 1543 dodata: /* XXX */ 1544 1545 /* 1546 * Process the segment text, merging it into the TCP sequencing queue, 1547 * and arranging for acknowledgment of receipt if necessary. 1548 * This process logically involves adjusting tp->rcv_wnd as data 1549 * is presented to the user (this happens in tcp_usrreq.c, 1550 * case PRU_RCVD). If a FIN has already been received on this 1551 * connection then we just ignore the text. 1552 */ 1553 if ((ti->ti_len || (tiflags&TH_FIN)) && 1554 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1555 TCP_REASS(tp, ti, m, so, tiflags); 1556 /* 1557 * Note the amount of data that peer has sent into 1558 * our window, in order to estimate the sender's 1559 * buffer size. 1560 */ 1561 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1562 } else { 1563 m_freem(m); 1564 tiflags &= ~TH_FIN; 1565 } 1566 1567 /* 1568 * If FIN is received ACK the FIN and let the user know 1569 * that the connection is closing. 1570 */ 1571 if (tiflags & TH_FIN) { 1572 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1573 socantrcvmore(so); 1574 /* 1575 * If connection is half-synchronized 1576 * (ie NEEDSYN flag on) then delay ACK, 1577 * so it may be piggybacked when SYN is sent. 1578 * Otherwise, since we received a FIN then no 1579 * more input can be expected, send ACK now. 1580 */ 1581 if (tp->t_flags & TF_NEEDSYN) 1582 tp->t_flags |= TF_DELACK; 1583 else 1584 tp->t_flags |= TF_ACKNOW; 1585 tp->rcv_nxt++; 1586 } 1587 switch (tp->t_state) { 1588 1589 /* 1590 * In SYN_RECEIVED and ESTABLISHED STATES 1591 * enter the CLOSE_WAIT state. 1592 */ 1593 case TCPS_SYN_RECEIVED: 1594 case TCPS_ESTABLISHED: 1595 tp->t_state = TCPS_CLOSE_WAIT; 1596 break; 1597 1598 /* 1599 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1600 * enter the CLOSING state. 1601 */ 1602 case TCPS_FIN_WAIT_1: 1603 tp->t_state = TCPS_CLOSING; 1604 break; 1605 1606 /* 1607 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1608 * starting the time-wait timer, turning off the other 1609 * standard timers. 1610 */ 1611 case TCPS_FIN_WAIT_2: 1612 tp->t_state = TCPS_TIME_WAIT; 1613 tcp_canceltimers(tp); 1614 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1615 if (tp->cc_recv != 0 && 1616 tp->t_duration < TCPTV_MSL) { 1617 tp->t_timer[TCPT_2MSL] = 1618 tp->t_rxtcur * TCPTV_TWTRUNC; 1619 /* For transaction client, force ACK now. */ 1620 tp->t_flags |= TF_ACKNOW; 1621 } 1622 else 1623 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1624 soisdisconnected(so); 1625 break; 1626 1627 /* 1628 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1629 */ 1630 case TCPS_TIME_WAIT: 1631 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1632 break; 1633 } 1634 } 1635 #ifdef TCPDEBUG 1636 if (so->so_options & SO_DEBUG) 1637 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1638 #endif 1639 1640 /* 1641 * Return any desired output. 1642 */ 1643 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1644 (void) tcp_output(tp); 1645 return; 1646 1647 dropafterack: 1648 /* 1649 * Generate an ACK dropping incoming segment if it occupies 1650 * sequence space, where the ACK reflects our state. 1651 */ 1652 if (tiflags & TH_RST) 1653 goto drop; 1654 #ifdef TCPDEBUG 1655 if (so->so_options & SO_DEBUG) 1656 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1657 #endif 1658 m_freem(m); 1659 tp->t_flags |= TF_ACKNOW; 1660 (void) tcp_output(tp); 1661 return; 1662 1663 dropwithreset: 1664 /* 1665 * Generate a RST, dropping incoming segment. 1666 * Make ACK acceptable to originator of segment. 1667 * Don't bother to respond if destination was broadcast/multicast. 1668 */ 1669 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1670 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1671 goto drop; 1672 #ifdef TCPDEBUG 1673 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1674 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1675 #endif 1676 if (tiflags & TH_ACK) 1677 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1678 else { 1679 if (tiflags & TH_SYN) 1680 ti->ti_len++; 1681 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1682 TH_RST|TH_ACK); 1683 } 1684 /* destroy temporarily created socket */ 1685 if (dropsocket) 1686 (void) soabort(so); 1687 return; 1688 1689 drop: 1690 /* 1691 * Drop space held by incoming segment and return. 1692 */ 1693 #ifdef TCPDEBUG 1694 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1695 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1696 #endif 1697 m_freem(m); 1698 /* destroy temporarily created socket */ 1699 if (dropsocket) 1700 (void) soabort(so); 1701 return; 1702 #ifndef TUBA_INCLUDE 1703 } 1704 1705 static void 1706 tcp_dooptions(tp, cp, cnt, ti, to) 1707 struct tcpcb *tp; 1708 u_char *cp; 1709 int cnt; 1710 struct tcpiphdr *ti; 1711 struct tcpopt *to; 1712 { 1713 u_short mss = 0; 1714 int opt, optlen; 1715 1716 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1717 opt = cp[0]; 1718 if (opt == TCPOPT_EOL) 1719 break; 1720 if (opt == TCPOPT_NOP) 1721 optlen = 1; 1722 else { 1723 optlen = cp[1]; 1724 if (optlen <= 0) 1725 break; 1726 } 1727 switch (opt) { 1728 1729 default: 1730 continue; 1731 1732 case TCPOPT_MAXSEG: 1733 if (optlen != TCPOLEN_MAXSEG) 1734 continue; 1735 if (!(ti->ti_flags & TH_SYN)) 1736 continue; 1737 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1738 NTOHS(mss); 1739 break; 1740 1741 case TCPOPT_WINDOW: 1742 if (optlen != TCPOLEN_WINDOW) 1743 continue; 1744 if (!(ti->ti_flags & TH_SYN)) 1745 continue; 1746 tp->t_flags |= TF_RCVD_SCALE; 1747 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1748 break; 1749 1750 case TCPOPT_TIMESTAMP: 1751 if (optlen != TCPOLEN_TIMESTAMP) 1752 continue; 1753 to->to_flag |= TOF_TS; 1754 bcopy((char *)cp + 2, 1755 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1756 NTOHL(to->to_tsval); 1757 bcopy((char *)cp + 6, 1758 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1759 NTOHL(to->to_tsecr); 1760 1761 /* 1762 * A timestamp received in a SYN makes 1763 * it ok to send timestamp requests and replies. 1764 */ 1765 if (ti->ti_flags & TH_SYN) { 1766 tp->t_flags |= TF_RCVD_TSTMP; 1767 tp->ts_recent = to->to_tsval; 1768 tp->ts_recent_age = tcp_now; 1769 } 1770 break; 1771 case TCPOPT_CC: 1772 if (optlen != TCPOLEN_CC) 1773 continue; 1774 to->to_flag |= TOF_CC; 1775 bcopy((char *)cp + 2, 1776 (char *)&to->to_cc, sizeof(to->to_cc)); 1777 NTOHL(to->to_cc); 1778 /* 1779 * A CC or CC.new option received in a SYN makes 1780 * it ok to send CC in subsequent segments. 1781 */ 1782 if (ti->ti_flags & TH_SYN) 1783 tp->t_flags |= TF_RCVD_CC; 1784 break; 1785 case TCPOPT_CCNEW: 1786 if (optlen != TCPOLEN_CC) 1787 continue; 1788 if (!(ti->ti_flags & TH_SYN)) 1789 continue; 1790 to->to_flag |= TOF_CCNEW; 1791 bcopy((char *)cp + 2, 1792 (char *)&to->to_cc, sizeof(to->to_cc)); 1793 NTOHL(to->to_cc); 1794 /* 1795 * A CC or CC.new option received in a SYN makes 1796 * it ok to send CC in subsequent segments. 1797 */ 1798 tp->t_flags |= TF_RCVD_CC; 1799 break; 1800 case TCPOPT_CCECHO: 1801 if (optlen != TCPOLEN_CC) 1802 continue; 1803 if (!(ti->ti_flags & TH_SYN)) 1804 continue; 1805 to->to_flag |= TOF_CCECHO; 1806 bcopy((char *)cp + 2, 1807 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1808 NTOHL(to->to_ccecho); 1809 break; 1810 } 1811 } 1812 if (ti->ti_flags & TH_SYN) 1813 tcp_mss(tp, mss); /* sets t_maxseg */ 1814 } 1815 1816 /* 1817 * Pull out of band byte out of a segment so 1818 * it doesn't appear in the user's data queue. 1819 * It is still reflected in the segment length for 1820 * sequencing purposes. 1821 */ 1822 static void 1823 tcp_pulloutofband(so, ti, m) 1824 struct socket *so; 1825 struct tcpiphdr *ti; 1826 register struct mbuf *m; 1827 { 1828 int cnt = ti->ti_urp - 1; 1829 1830 while (cnt >= 0) { 1831 if (m->m_len > cnt) { 1832 char *cp = mtod(m, caddr_t) + cnt; 1833 struct tcpcb *tp = sototcpcb(so); 1834 1835 tp->t_iobc = *cp; 1836 tp->t_oobflags |= TCPOOB_HAVEDATA; 1837 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1838 m->m_len--; 1839 return; 1840 } 1841 cnt -= m->m_len; 1842 m = m->m_next; 1843 if (m == 0) 1844 break; 1845 } 1846 panic("tcp_pulloutofband"); 1847 } 1848 1849 /* 1850 * Collect new round-trip time estimate 1851 * and update averages and current timeout. 1852 */ 1853 static void 1854 tcp_xmit_timer(tp, rtt) 1855 register struct tcpcb *tp; 1856 short rtt; 1857 { 1858 register int delta; 1859 1860 tcpstat.tcps_rttupdated++; 1861 tp->t_rttupdated++; 1862 if (tp->t_srtt != 0) { 1863 /* 1864 * srtt is stored as fixed point with 5 bits after the 1865 * binary point (i.e., scaled by 8). The following magic 1866 * is equivalent to the smoothing algorithm in rfc793 with 1867 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1868 * point). Adjust rtt to origin 0. 1869 */ 1870 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1871 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1872 1873 if ((tp->t_srtt += delta) <= 0) 1874 tp->t_srtt = 1; 1875 1876 /* 1877 * We accumulate a smoothed rtt variance (actually, a 1878 * smoothed mean difference), then set the retransmit 1879 * timer to smoothed rtt + 4 times the smoothed variance. 1880 * rttvar is stored as fixed point with 4 bits after the 1881 * binary point (scaled by 16). The following is 1882 * equivalent to rfc793 smoothing with an alpha of .75 1883 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1884 * rfc793's wired-in beta. 1885 */ 1886 if (delta < 0) 1887 delta = -delta; 1888 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1889 if ((tp->t_rttvar += delta) <= 0) 1890 tp->t_rttvar = 1; 1891 } else { 1892 /* 1893 * No rtt measurement yet - use the unsmoothed rtt. 1894 * Set the variance to half the rtt (so our first 1895 * retransmit happens at 3*rtt). 1896 */ 1897 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1898 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1899 } 1900 tp->t_rtt = 0; 1901 tp->t_rxtshift = 0; 1902 1903 /* 1904 * the retransmit should happen at rtt + 4 * rttvar. 1905 * Because of the way we do the smoothing, srtt and rttvar 1906 * will each average +1/2 tick of bias. When we compute 1907 * the retransmit timer, we want 1/2 tick of rounding and 1908 * 1 extra tick because of +-1/2 tick uncertainty in the 1909 * firing of the timer. The bias will give us exactly the 1910 * 1.5 tick we need. But, because the bias is 1911 * statistical, we have to test that we don't drop below 1912 * the minimum feasible timer (which is 2 ticks). 1913 */ 1914 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1915 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1916 1917 /* 1918 * We received an ack for a packet that wasn't retransmitted; 1919 * it is probably safe to discard any error indications we've 1920 * received recently. This isn't quite right, but close enough 1921 * for now (a route might have failed after we sent a segment, 1922 * and the return path might not be symmetrical). 1923 */ 1924 tp->t_softerror = 0; 1925 } 1926 1927 /* 1928 * Determine a reasonable value for maxseg size. 1929 * If the route is known, check route for mtu. 1930 * If none, use an mss that can be handled on the outgoing 1931 * interface without forcing IP to fragment; if bigger than 1932 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1933 * to utilize large mbufs. If no route is found, route has no mtu, 1934 * or the destination isn't local, use a default, hopefully conservative 1935 * size (usually 512 or the default IP max size, but no more than the mtu 1936 * of the interface), as we can't discover anything about intervening 1937 * gateways or networks. We also initialize the congestion/slow start 1938 * window to be a single segment if the destination isn't local. 1939 * While looking at the routing entry, we also initialize other path-dependent 1940 * parameters from pre-set or cached values in the routing entry. 1941 * 1942 * Also take into account the space needed for options that we 1943 * send regularly. Make maxseg shorter by that amount to assure 1944 * that we can send maxseg amount of data even when the options 1945 * are present. Store the upper limit of the length of options plus 1946 * data in maxopd. 1947 * 1948 * NOTE that this routine is only called when we process an incoming 1949 * segment, for outgoing segments only tcp_mssopt is called. 1950 * 1951 * In case of T/TCP, we call this routine during implicit connection 1952 * setup as well (offer = -1), to initialize maxseg from the cached 1953 * MSS of our peer. 1954 */ 1955 void 1956 tcp_mss(tp, offer) 1957 struct tcpcb *tp; 1958 int offer; 1959 { 1960 register struct rtentry *rt; 1961 struct ifnet *ifp; 1962 register int rtt, mss; 1963 u_long bufsize; 1964 struct inpcb *inp; 1965 struct socket *so; 1966 struct rmxp_tao *taop; 1967 int origoffer = offer; 1968 1969 inp = tp->t_inpcb; 1970 if ((rt = tcp_rtlookup(inp)) == NULL) { 1971 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1972 return; 1973 } 1974 ifp = rt->rt_ifp; 1975 so = inp->inp_socket; 1976 1977 taop = rmx_taop(rt->rt_rmx); 1978 /* 1979 * Offer == -1 means that we didn't receive SYN yet, 1980 * use cached value in that case; 1981 */ 1982 if (offer == -1) 1983 offer = taop->tao_mssopt; 1984 /* 1985 * Offer == 0 means that there was no MSS on the SYN segment, 1986 * in this case we use tcp_mssdflt. 1987 */ 1988 if (offer == 0) 1989 offer = tcp_mssdflt; 1990 else 1991 /* 1992 * Sanity check: make sure that maxopd will be large 1993 * enough to allow some data on segments even is the 1994 * all the option space is used (40bytes). Otherwise 1995 * funny things may happen in tcp_output. 1996 */ 1997 offer = max(offer, 64); 1998 taop->tao_mssopt = offer; 1999 2000 /* 2001 * While we're here, check if there's an initial rtt 2002 * or rttvar. Convert from the route-table units 2003 * to scaled multiples of the slow timeout timer. 2004 */ 2005 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2006 /* 2007 * XXX the lock bit for RTT indicates that the value 2008 * is also a minimum value; this is subject to time. 2009 */ 2010 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2011 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2012 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2013 tcpstat.tcps_usedrtt++; 2014 if (rt->rt_rmx.rmx_rttvar) { 2015 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2016 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2017 tcpstat.tcps_usedrttvar++; 2018 } else { 2019 /* default variation is +- 1 rtt */ 2020 tp->t_rttvar = 2021 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2022 } 2023 TCPT_RANGESET(tp->t_rxtcur, 2024 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2025 tp->t_rttmin, TCPTV_REXMTMAX); 2026 } 2027 /* 2028 * if there's an mtu associated with the route, use it 2029 */ 2030 if (rt->rt_rmx.rmx_mtu) 2031 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2032 else 2033 { 2034 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2035 if (!in_localaddr(inp->inp_faddr)) 2036 mss = min(mss, tcp_mssdflt); 2037 } 2038 mss = min(mss, offer); 2039 /* 2040 * maxopd stores the maximum length of data AND options 2041 * in a segment; maxseg is the amount of data in a normal 2042 * segment. We need to store this value (maxopd) apart 2043 * from maxseg, because now every segment carries options 2044 * and thus we normally have somewhat less data in segments. 2045 */ 2046 tp->t_maxopd = mss; 2047 2048 /* 2049 * In case of T/TCP, origoffer==-1 indicates, that no segments 2050 * were received yet. In this case we just guess, otherwise 2051 * we do the same as before T/TCP. 2052 */ 2053 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2054 (origoffer == -1 || 2055 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2056 mss -= TCPOLEN_TSTAMP_APPA; 2057 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2058 (origoffer == -1 || 2059 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2060 mss -= TCPOLEN_CC_APPA; 2061 2062 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2063 if (mss > MCLBYTES) 2064 mss &= ~(MCLBYTES-1); 2065 #else 2066 if (mss > MCLBYTES) 2067 mss = mss / MCLBYTES * MCLBYTES; 2068 #endif 2069 /* 2070 * If there's a pipesize, change the socket buffer 2071 * to that size. Make the socket buffers an integral 2072 * number of mss units; if the mss is larger than 2073 * the socket buffer, decrease the mss. 2074 */ 2075 #ifdef RTV_SPIPE 2076 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2077 #endif 2078 bufsize = so->so_snd.sb_hiwat; 2079 if (bufsize < mss) 2080 mss = bufsize; 2081 else { 2082 bufsize = roundup(bufsize, mss); 2083 if (bufsize > sb_max) 2084 bufsize = sb_max; 2085 (void)sbreserve(&so->so_snd, bufsize); 2086 } 2087 tp->t_maxseg = mss; 2088 2089 #ifdef RTV_RPIPE 2090 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2091 #endif 2092 bufsize = so->so_rcv.sb_hiwat; 2093 if (bufsize > mss) { 2094 bufsize = roundup(bufsize, mss); 2095 if (bufsize > sb_max) 2096 bufsize = sb_max; 2097 (void)sbreserve(&so->so_rcv, bufsize); 2098 } 2099 /* 2100 * Don't force slow-start on local network. 2101 */ 2102 if (!in_localaddr(inp->inp_faddr)) 2103 tp->snd_cwnd = mss; 2104 2105 if (rt->rt_rmx.rmx_ssthresh) { 2106 /* 2107 * There's some sort of gateway or interface 2108 * buffer limit on the path. Use this to set 2109 * the slow start threshhold, but set the 2110 * threshold to no less than 2*mss. 2111 */ 2112 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2113 tcpstat.tcps_usedssthresh++; 2114 } 2115 } 2116 2117 /* 2118 * Determine the MSS option to send on an outgoing SYN. 2119 */ 2120 int 2121 tcp_mssopt(tp) 2122 struct tcpcb *tp; 2123 { 2124 struct rtentry *rt; 2125 2126 rt = tcp_rtlookup(tp->t_inpcb); 2127 if (rt == NULL) 2128 return tcp_mssdflt; 2129 2130 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2131 } 2132 #endif /* TUBA_INCLUDE */ 2133